ReMIND: An AI Assistant for Neuroimaging Analysis
Researchers in the Kolachalama Laboratory at Boston University have recently developed ReMIND, a tool powered by artificial intelligence (AI) to aid clinicians with the growing demand of interpreting and reporting on large quantities of multi-sequence brain scans.
With the increasing incidence of neurological disorders, clinicians are currently facing a higher demand of patients requiring MRIs. These capacity constraints can put pressures on the quality and accessibility of care for patients, risking delayed diagnosis. ReMIND offers a unique opportunity to streamline and advance these processes, with the goal of benefiting both patients and clinicians.
ReMIND, which stands for “Radiology–encoded Multimodal Interpretation for Neurological Disorders”, is a vision-language model that leverages computer vision and natural language processing to interpret multi-sequence brain MRI scans and draft radiology reports. The model generates an initial report after reviewing scans, and then it verifies its output by checking that each claim is substantiated by the necessary, corresponding MRI sequence.
The objective for ReMIND is to assist clinicians and researchers in their process. The outputs are designed to be preliminary report drafts that are then reviewed, edited, and finalized by clinicians or subject matter experts. This oversight from qualified experts is critical for ReMIND’s implementation.
ReMIND was trained on deidentified data from over tens of thousands of patient visits, which included close to a million MRI sequences from a diverse cohort of international datasets. To ensure data privacy and security, no patient data is stored by the program’s servers after the session ends.
“ReMIND matters because it moves us toward AI systems that can interpret complex brain MRI studies in a clinically grounded way, helping clinicians and researchers generate more consistent, faithful, and useful imaging summaries for patient care,” said Diala Lteif who led the development of ReMIND and recently defended her PhD at Boston University.
The developers of ReMIND aim to continue evaluating the model’s performance in realistic clinical and research settings. The team is actively working with several clinicians to assess the accuracy and usefulness of ReMIND, while refining the model and web interface to make the tool easier to use.
This technology is currently a research prototype and is available to the Boston University community using BU login credentials at remind.bu.edu. For those at external institutions or organizations looking to try ReMIND, access is available upon request. This tool is not approved for clinical diagnosis.
Learn more about the development of ReMIND by reading the catalyst research paper.